Regression Analysis based on User Characteristics

from pycaret.regression import *

# Data Load
df_user = pd.read_csv("./preprocessed_data/df2021_user.csv")

# Delete rows where the spending amount is zero
df_user_nz = df_user[df_user["next_paid"] != 0]

# Check the top 5 rows
df_user_nz.head()

# Convert next month's charge amount to categories
def convert_paid(x):
    if (x >= 0) and (x <= 1000):
        x = 1
    elif (x > 1000) and (x <= 5000):
        x = 2
    elif (x > 5000) and (x <= 10000):
        x = 3
    elif (x > 10000) and (x <= 20000):
        x = 4
    elif (x > 20000) and (x <= 30000):
        x = 5
    elif (x > 30000) and (x <= 40000):
        x = 6
    elif (x > 40000) and (x <= 50000):
        x = 7
    elif (x > 50000) and (x <= 100000):
        x = 8
    else:
        x = 9
    
    return x

df_user_nz["current_paid"] = df_user_nz["current_paid"].apply(convert_paid)
df_user_nz["next_paid"] = df_user_nz["next_paid"].apply(convert_paid)

# Data Split
features = df_user_nz.drop(["next_attendance", "next_paid"], axis = 1)
label = df_user_nz["next_attendance"]

features_train, features_test, label_train, label_test = train_test_split(features, label, test_size = 0.2, random_state = 2023, shuffle = True, stratify = label)

df_train = pd.concat([features_train, label_train], axis = 1)

reg = setup(data = df_train, target = "next_attendance", train_size = 0.8)

# Top 3 models
top3_models = compare_models(round = 3, sort = 'RMSE', n_select = 3)

# Create model
gbr_model = create_model('gbr', cross_validation=True)

# Hyperparameter tuning
tuned_gbr_model = tune_model(gbr_model, optimize = "RMSE")

# Finalize Model Selection
final_gbr_model = finalize_model(tuned_gbr_model)

prediction = predict_model(final_gbr_model, data = features_test)

# Accuracy (not precise due to it being a regression)
pred_acc = (prediction["prediction_label"] == label_test).sum() / len(label_test)
pred_acc * 100

6.559263521288838

prediction["prediction_label"] = np.round(prediction["prediction_label"], 0)

# Based on the prediction["prediction_label"] column & label_test column
predict_result = pd.concat([prediction["prediction_label"], label_test], axis = 1)

# Extract into a CSV file to verify accuracy
predict_result.to_csv("./regression_result_gbr.csv", index=False, encoding="utf-8-sig")